Politics and The Tools of Artificial Intelligence (page 1 of 2)
By Denny Rock

In 1776, Thomas Paine sold a half-million copies of Common Sense to a nation only three times that size. His goal was to awaken that "human mass of sense lying in a dormant state" to fight for political change. Today, tremendous changes in advanced computing technologies are giving rise to a similar challenge of democratic empowerment. A number of major political figures and trends are already employing these resources:

How will computer technology affect future politics? This short survey is not meant to be a soapbox for any particular political viewpoint. Instead, it stumps for the application of Artificial Intelligence (AI) technology to expand public involvement with information-driven politics, the politics of knowledge, not necessarily the politics of winning elections. I will point to some potential AI contributions: political models, tools to search for and assess political facts, tools to frame political concepts, and also tools to expand electronic discussion.

All Models Are Local

Computer models and simulation are needed to track even the roughest outlines of the increasingly complex political landscapes and to understand the dynamics of the underlying power realities. Political models achieve two goals: They locate candidates in what R. Joslyn calls "issue space" by analyzing the content of candidate appeals and making informed guesses about candidates' programmatic behavior once in office. They also attempt to understand the role of partisanship for example, the primary win by former Illinois Representative Dan Rostenkowski, even though he later lost his seat, was not influenced by issues as much as by perceived steadfastness and party loyalty.

One approach to modeling the behavior of political parties uses the artificial adaptive agent structures developed by John Holland and John Miller in the Echo class of models for complex adaptive systems. Echo models let researchers explore the relationship between optimization and adaptation and test hypotheses about the underlying environment. Echo's ability to represent the "unconscious internal models" might be useful for modeling the political thought processes of citizens. Likewise, Echo's ability to represent "aggregate behavior" might be useful for modeling the organizational evolution of a political party itself. Echo is available via anonymous ftp to ftp.santafe.edu for the file /pub/Users/terry/echo/Echo-1.0.tar.Z).

Smart Whistles and Watchdogs

AI tools for knowledge discovery are used to detect patterns of fraud in credit card and business applications. Can similar approaches be used in the political and governmental domains?

Taxpayers Against Fraud (TAF), a Washington, D.C.-based nonprofit organization, has recovered more than $588 million for the U.S. government since 1986. TAF uses the whistleblower' law to uncover fraud. This law originated when Abraham Lincoln cracked down on war profiteers who filled musket crates with sawdust and sold the same horses to the cavalry time after time.

Lisa Hovelson, executive director of TAF, says that computers have been used only to calculate damages after fraud details are known, not at the front end for data discovery or analysis, for which TAF essentially has relied on inside persons. "We have discussed and support the need for such AI capability, but it is still in the future for us," says Hovelson. An example of U.S. government interagency exchange of information, where data correlation is required, suggests Hovelson, is the IRS and the Department of Education for defaulted student loans. Another example is the Department of Customs and the duties paid on products coming into the United States, compared to the prices charged to the government.

Yet another potential application involves watchdogs for vote fraud. A recent case involving a close election loss for the Pennsylvania State Senate by Republican Bruce Marks kept the Philadelphia news media humming for months. The election had slipped by the watch of the nonpartisan group, which manually inspects ballots and allegations of election impropriety. A pattern of ballot fraud and forgery was detected after citizens protested that their names were on erroneous absentee ballots. A Philadelphia Inquirer editorial called for "modernizing voter registration information by computerization including digitizing signatures."

Forensic Linguistics Reliable information is essential for a free-thinking public to arrive at opinions. New computer applications can assist in the related functions of news understanding, text retrieval, and the acknowledgment of bias or intentional ambiguity. Such applications could assist journalists, as well as citizens.

The Arlington,Va. based Advanced Research Projects Agency (ARPA) has sponsored a series of Message Understanding Conference (MUC) competitions.The goal in MUC 3 concerned the extraction of information from news articles about the topic of terrorism. MUC solutions have ranged from in-depth natural-language understanding capabilities to skimming techniques that aim to avoid the knowledge-engineering bottleneck associated with many text-processing systems.

Mainstream journalism in the wire services the primary source for most of the 1,800 daily newspaper, 11,000 radio, and 2,000 TV stations in North America is characterized generally by neutrality and balance. Exceptions exist, and the detection of linguistic bias in the news media is very important. A few of the news services that focus on the exceptions include FAIR (Fairness & Accuracy In Reporting), LOOT (Lies Of Our Times, Institute of Media Analysis), and Critical Intelligence (Boardroom Inc.), all based in New York.

Fuzzy Detective Tools

L. Bennett suggests that implicit handling of policy information by the news media would not be a problem for democracy if members of the public approached the news as detectives, looking for hidden clues upon which to build their understanding about a situation. Libraries already use electronic-search capabilities for information filtering, document location, and fact extraction. Software tools that achieve these tasks include Gopher, Wide Area Information Servers, Archie, and AppleSearch. While these first-generation tools have been limited by keyword requirements, the commercial development of fuzzy search' capabilities in a few expensive tools is a harbinger.

One fuzzy search' tool vendor is Excalibur Technologies Inc. (San Diego, Calif.). Excalibur's document-retrieval products have migrated to client/server architectures and will be offered by late 1994 as an unbundled set of advanced programming tools for embedded applications. Metrics given by Excalibur include search 200,000 pages of text in ten seconds, learn new input data at a rate of five megabytes in 160 seconds, and create index memories a third of the size of the original text.While Excalibur's pattern-recognition tools have been applied to text and picture images, multimedia applications with digital data of voice or video are yet to be explored in this domain.

Unlike many traditional search-and-retrieval systems that discard certain words such as "the," Excalibur's approach can search on concepts or every single word. For example, "The" is a common Vietnamese name and is featured prominently in many Defense documents of the Vietnam War era. The Library of Congress uses Excalibur's tool to scan in Spanish-language law journals from around the world. The Defense Intelligence Agency's Counter-Drug Directorate uses this tool to scan in articles from Spanish newspapers and search for words and images. The U.S. Department of Defense's Decision Systems Management Agency uses this tool to process records from the former Soviet Union, searching for clues related to U.S. prisoners of war. More >>